https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97286
Jan Misiak <fijam7 at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |fijam7 at gmail dot com
--- Comment #20 from Jan Misiak <fijam7 at gmail dot com> ---
This is still the case as of GCC 16.2. I found it landing on bug 94663, which
is marked as a duplicate of this one. Here is a small test case (and a
workaround with V16SI accumulators): https://godbolt.org/z/119v8K95h
Without the workaround, this costs me around 4% in an AVX-512 kernel code on
Cascade Lake compared to a clang build which produces a copy-free loop.
